I Know We're in an AI Bubble Because Nobody Wants Me
12 days ago
- #deep learning
- #AI efficiency
- #hardware vs software
- The author first got into deep learning in 2012 with the release of AlexNet, which was revolutionary for their work at Jetpac.
- Faced with the challenge of applying deep learning models to billions of photos, the author developed the Jetpac framework to run models efficiently on low-cost hardware.
- The author's passion for efficiency stems from their background in coding and optimizing systems, dating back to the 80’s demo scene and 90’s game engine development.
- Optimization is described as an emotional and rewarding process, involving deep understanding of hardware, software, and teamwork to achieve better performance.
- The author criticizes the current AI bubble, where billions are spent on hardware like GPUs, while software optimization and efficiency improvements are undervalued.
- Companies and startups prioritize hardware spending as a signaling mechanism to investors and the market, rather than focusing on cost-effective software solutions.
- The author predicts that the current trend of heavy investment in expensive hardware is unsustainable, drawing parallels to the dot com boom and the eventual shift to cheaper, scalable solutions.
- Despite their predictions, the author acknowledges that Nvidia's valuation has continued to rise, indicating the complexity of market trends.